• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	fs/proc/vmcore.c Interface for accessing the crash
4  * 				 dump from the system's previous life.
5  * 	Heavily borrowed from fs/proc/kcore.c
6  *	Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
7  *	Copyright (C) IBM Corporation, 2004. All rights reserved
8  *
9  */
10 
11 #include <linux/mm.h>
12 #include <linux/kcore.h>
13 #include <linux/user.h>
14 #include <linux/elf.h>
15 #include <linux/elfcore.h>
16 #include <linux/export.h>
17 #include <linux/slab.h>
18 #include <linux/highmem.h>
19 #include <linux/printk.h>
20 #include <linux/memblock.h>
21 #include <linux/init.h>
22 #include <linux/crash_dump.h>
23 #include <linux/list.h>
24 #include <linux/moduleparam.h>
25 #include <linux/mutex.h>
26 #include <linux/vmalloc.h>
27 #include <linux/pagemap.h>
28 #include <linux/uaccess.h>
29 #include <linux/mem_encrypt.h>
30 #include <asm/io.h>
31 #include "internal.h"
32 
33 /* List representing chunks of contiguous memory areas and their offsets in
34  * vmcore file.
35  */
36 static LIST_HEAD(vmcore_list);
37 
38 /* Stores the pointer to the buffer containing kernel elf core headers. */
39 static char *elfcorebuf;
40 static size_t elfcorebuf_sz;
41 static size_t elfcorebuf_sz_orig;
42 
43 static char *elfnotes_buf;
44 static size_t elfnotes_sz;
45 /* Size of all notes minus the device dump notes */
46 static size_t elfnotes_orig_sz;
47 
48 /* Total size of vmcore file. */
49 static u64 vmcore_size;
50 
51 static struct proc_dir_entry *proc_vmcore;
52 
53 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
54 /* Device Dump list and mutex to synchronize access to list */
55 static LIST_HEAD(vmcoredd_list);
56 static DEFINE_MUTEX(vmcoredd_mutex);
57 
58 static bool vmcoredd_disabled;
59 core_param(novmcoredd, vmcoredd_disabled, bool, 0);
60 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
61 
62 /* Device Dump Size */
63 static size_t vmcoredd_orig_sz;
64 
65 /*
66  * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
67  * The called function has to take care of module refcounting.
68  */
69 static int (*oldmem_pfn_is_ram)(unsigned long pfn);
70 
register_oldmem_pfn_is_ram(int (* fn)(unsigned long pfn))71 int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn))
72 {
73 	if (oldmem_pfn_is_ram)
74 		return -EBUSY;
75 	oldmem_pfn_is_ram = fn;
76 	return 0;
77 }
78 EXPORT_SYMBOL_GPL(register_oldmem_pfn_is_ram);
79 
unregister_oldmem_pfn_is_ram(void)80 void unregister_oldmem_pfn_is_ram(void)
81 {
82 	oldmem_pfn_is_ram = NULL;
83 	wmb();
84 }
85 EXPORT_SYMBOL_GPL(unregister_oldmem_pfn_is_ram);
86 
pfn_is_ram(unsigned long pfn)87 static int pfn_is_ram(unsigned long pfn)
88 {
89 	int (*fn)(unsigned long pfn);
90 	/* pfn is ram unless fn() checks pagetype */
91 	int ret = 1;
92 
93 	/*
94 	 * Ask hypervisor if the pfn is really ram.
95 	 * A ballooned page contains no data and reading from such a page
96 	 * will cause high load in the hypervisor.
97 	 */
98 	fn = oldmem_pfn_is_ram;
99 	if (fn)
100 		ret = fn(pfn);
101 
102 	return ret;
103 }
104 
105 /* Reads a page from the oldmem device from given offset. */
read_from_oldmem(char * buf,size_t count,u64 * ppos,int userbuf,bool encrypted)106 ssize_t read_from_oldmem(char *buf, size_t count,
107 			 u64 *ppos, int userbuf,
108 			 bool encrypted)
109 {
110 	unsigned long pfn, offset;
111 	size_t nr_bytes;
112 	ssize_t read = 0, tmp;
113 
114 	if (!count)
115 		return 0;
116 
117 	offset = (unsigned long)(*ppos % PAGE_SIZE);
118 	pfn = (unsigned long)(*ppos / PAGE_SIZE);
119 
120 	do {
121 		if (count > (PAGE_SIZE - offset))
122 			nr_bytes = PAGE_SIZE - offset;
123 		else
124 			nr_bytes = count;
125 
126 		/* If pfn is not ram, return zeros for sparse dump files */
127 		if (pfn_is_ram(pfn) == 0) {
128 			tmp = 0;
129 			if (!userbuf)
130 				memset(buf, 0, nr_bytes);
131 			else if (clear_user(buf, nr_bytes))
132 				tmp = -EFAULT;
133 		} else {
134 			if (encrypted)
135 				tmp = copy_oldmem_page_encrypted(pfn, buf,
136 								 nr_bytes,
137 								 offset,
138 								 userbuf);
139 			else
140 				tmp = copy_oldmem_page(pfn, buf, nr_bytes,
141 						       offset, userbuf);
142 		}
143 		if (tmp < 0)
144 			return tmp;
145 
146 		*ppos += nr_bytes;
147 		count -= nr_bytes;
148 		buf += nr_bytes;
149 		read += nr_bytes;
150 		++pfn;
151 		offset = 0;
152 	} while (count);
153 
154 	return read;
155 }
156 
157 /*
158  * Architectures may override this function to allocate ELF header in 2nd kernel
159  */
elfcorehdr_alloc(unsigned long long * addr,unsigned long long * size)160 int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
161 {
162 	return 0;
163 }
164 
165 /*
166  * Architectures may override this function to free header
167  */
elfcorehdr_free(unsigned long long addr)168 void __weak elfcorehdr_free(unsigned long long addr)
169 {}
170 
171 /*
172  * Architectures may override this function to read from ELF header
173  */
elfcorehdr_read(char * buf,size_t count,u64 * ppos)174 ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
175 {
176 	return read_from_oldmem(buf, count, ppos, 0, false);
177 }
178 
179 /*
180  * Architectures may override this function to read from notes sections
181  */
elfcorehdr_read_notes(char * buf,size_t count,u64 * ppos)182 ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
183 {
184 	return read_from_oldmem(buf, count, ppos, 0, mem_encrypt_active());
185 }
186 
187 /*
188  * Architectures may override this function to map oldmem
189  */
remap_oldmem_pfn_range(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)190 int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
191 				  unsigned long from, unsigned long pfn,
192 				  unsigned long size, pgprot_t prot)
193 {
194 	prot = pgprot_encrypted(prot);
195 	return remap_pfn_range(vma, from, pfn, size, prot);
196 }
197 
198 /*
199  * Architectures which support memory encryption override this.
200  */
201 ssize_t __weak
copy_oldmem_page_encrypted(unsigned long pfn,char * buf,size_t csize,unsigned long offset,int userbuf)202 copy_oldmem_page_encrypted(unsigned long pfn, char *buf, size_t csize,
203 			   unsigned long offset, int userbuf)
204 {
205 	return copy_oldmem_page(pfn, buf, csize, offset, userbuf);
206 }
207 
208 /*
209  * Copy to either kernel or user space
210  */
copy_to(void * target,void * src,size_t size,int userbuf)211 static int copy_to(void *target, void *src, size_t size, int userbuf)
212 {
213 	if (userbuf) {
214 		if (copy_to_user((char __user *) target, src, size))
215 			return -EFAULT;
216 	} else {
217 		memcpy(target, src, size);
218 	}
219 	return 0;
220 }
221 
222 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
vmcoredd_copy_dumps(void * dst,u64 start,size_t size,int userbuf)223 static int vmcoredd_copy_dumps(void *dst, u64 start, size_t size, int userbuf)
224 {
225 	struct vmcoredd_node *dump;
226 	u64 offset = 0;
227 	int ret = 0;
228 	size_t tsz;
229 	char *buf;
230 
231 	mutex_lock(&vmcoredd_mutex);
232 	list_for_each_entry(dump, &vmcoredd_list, list) {
233 		if (start < offset + dump->size) {
234 			tsz = min(offset + (u64)dump->size - start, (u64)size);
235 			buf = dump->buf + start - offset;
236 			if (copy_to(dst, buf, tsz, userbuf)) {
237 				ret = -EFAULT;
238 				goto out_unlock;
239 			}
240 
241 			size -= tsz;
242 			start += tsz;
243 			dst += tsz;
244 
245 			/* Leave now if buffer filled already */
246 			if (!size)
247 				goto out_unlock;
248 		}
249 		offset += dump->size;
250 	}
251 
252 out_unlock:
253 	mutex_unlock(&vmcoredd_mutex);
254 	return ret;
255 }
256 
257 #ifdef CONFIG_MMU
vmcoredd_mmap_dumps(struct vm_area_struct * vma,unsigned long dst,u64 start,size_t size)258 static int vmcoredd_mmap_dumps(struct vm_area_struct *vma, unsigned long dst,
259 			       u64 start, size_t size)
260 {
261 	struct vmcoredd_node *dump;
262 	u64 offset = 0;
263 	int ret = 0;
264 	size_t tsz;
265 	char *buf;
266 
267 	mutex_lock(&vmcoredd_mutex);
268 	list_for_each_entry(dump, &vmcoredd_list, list) {
269 		if (start < offset + dump->size) {
270 			tsz = min(offset + (u64)dump->size - start, (u64)size);
271 			buf = dump->buf + start - offset;
272 			if (remap_vmalloc_range_partial(vma, dst, buf, 0,
273 							tsz)) {
274 				ret = -EFAULT;
275 				goto out_unlock;
276 			}
277 
278 			size -= tsz;
279 			start += tsz;
280 			dst += tsz;
281 
282 			/* Leave now if buffer filled already */
283 			if (!size)
284 				goto out_unlock;
285 		}
286 		offset += dump->size;
287 	}
288 
289 out_unlock:
290 	mutex_unlock(&vmcoredd_mutex);
291 	return ret;
292 }
293 #endif /* CONFIG_MMU */
294 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
295 
296 /* Read from the ELF header and then the crash dump. On error, negative value is
297  * returned otherwise number of bytes read are returned.
298  */
__read_vmcore(char * buffer,size_t buflen,loff_t * fpos,int userbuf)299 static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
300 			     int userbuf)
301 {
302 	ssize_t acc = 0, tmp;
303 	size_t tsz;
304 	u64 start;
305 	struct vmcore *m = NULL;
306 
307 	if (buflen == 0 || *fpos >= vmcore_size)
308 		return 0;
309 
310 	/* trim buflen to not go beyond EOF */
311 	if (buflen > vmcore_size - *fpos)
312 		buflen = vmcore_size - *fpos;
313 
314 	/* Read ELF core header */
315 	if (*fpos < elfcorebuf_sz) {
316 		tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
317 		if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
318 			return -EFAULT;
319 		buflen -= tsz;
320 		*fpos += tsz;
321 		buffer += tsz;
322 		acc += tsz;
323 
324 		/* leave now if filled buffer already */
325 		if (buflen == 0)
326 			return acc;
327 	}
328 
329 	/* Read Elf note segment */
330 	if (*fpos < elfcorebuf_sz + elfnotes_sz) {
331 		void *kaddr;
332 
333 		/* We add device dumps before other elf notes because the
334 		 * other elf notes may not fill the elf notes buffer
335 		 * completely and we will end up with zero-filled data
336 		 * between the elf notes and the device dumps. Tools will
337 		 * then try to decode this zero-filled data as valid notes
338 		 * and we don't want that. Hence, adding device dumps before
339 		 * the other elf notes ensure that zero-filled data can be
340 		 * avoided.
341 		 */
342 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
343 		/* Read device dumps */
344 		if (*fpos < elfcorebuf_sz + vmcoredd_orig_sz) {
345 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
346 				  (size_t)*fpos, buflen);
347 			start = *fpos - elfcorebuf_sz;
348 			if (vmcoredd_copy_dumps(buffer, start, tsz, userbuf))
349 				return -EFAULT;
350 
351 			buflen -= tsz;
352 			*fpos += tsz;
353 			buffer += tsz;
354 			acc += tsz;
355 
356 			/* leave now if filled buffer already */
357 			if (!buflen)
358 				return acc;
359 		}
360 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
361 
362 		/* Read remaining elf notes */
363 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
364 		kaddr = elfnotes_buf + *fpos - elfcorebuf_sz - vmcoredd_orig_sz;
365 		if (copy_to(buffer, kaddr, tsz, userbuf))
366 			return -EFAULT;
367 
368 		buflen -= tsz;
369 		*fpos += tsz;
370 		buffer += tsz;
371 		acc += tsz;
372 
373 		/* leave now if filled buffer already */
374 		if (buflen == 0)
375 			return acc;
376 	}
377 
378 	list_for_each_entry(m, &vmcore_list, list) {
379 		if (*fpos < m->offset + m->size) {
380 			tsz = (size_t)min_t(unsigned long long,
381 					    m->offset + m->size - *fpos,
382 					    buflen);
383 			start = m->paddr + *fpos - m->offset;
384 			tmp = read_from_oldmem(buffer, tsz, &start,
385 					       userbuf, mem_encrypt_active());
386 			if (tmp < 0)
387 				return tmp;
388 			buflen -= tsz;
389 			*fpos += tsz;
390 			buffer += tsz;
391 			acc += tsz;
392 
393 			/* leave now if filled buffer already */
394 			if (buflen == 0)
395 				return acc;
396 		}
397 	}
398 
399 	return acc;
400 }
401 
read_vmcore(struct file * file,char __user * buffer,size_t buflen,loff_t * fpos)402 static ssize_t read_vmcore(struct file *file, char __user *buffer,
403 			   size_t buflen, loff_t *fpos)
404 {
405 	return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
406 }
407 
408 /*
409  * The vmcore fault handler uses the page cache and fills data using the
410  * standard __vmcore_read() function.
411  *
412  * On s390 the fault handler is used for memory regions that can't be mapped
413  * directly with remap_pfn_range().
414  */
mmap_vmcore_fault(struct vm_fault * vmf)415 static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
416 {
417 #ifdef CONFIG_S390
418 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
419 	pgoff_t index = vmf->pgoff;
420 	struct page *page;
421 	loff_t offset;
422 	char *buf;
423 	int rc;
424 
425 	page = find_or_create_page(mapping, index, GFP_KERNEL);
426 	if (!page)
427 		return VM_FAULT_OOM;
428 	if (!PageUptodate(page)) {
429 		offset = (loff_t) index << PAGE_SHIFT;
430 		buf = __va((page_to_pfn(page) << PAGE_SHIFT));
431 		rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
432 		if (rc < 0) {
433 			unlock_page(page);
434 			put_page(page);
435 			return vmf_error(rc);
436 		}
437 		SetPageUptodate(page);
438 	}
439 	unlock_page(page);
440 	vmf->page = page;
441 	return 0;
442 #else
443 	return VM_FAULT_SIGBUS;
444 #endif
445 }
446 
447 static const struct vm_operations_struct vmcore_mmap_ops = {
448 	.fault = mmap_vmcore_fault,
449 };
450 
451 /**
452  * vmcore_alloc_buf - allocate buffer in vmalloc memory
453  * @sizez: size of buffer
454  *
455  * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
456  * the buffer to user-space by means of remap_vmalloc_range().
457  *
458  * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
459  * disabled and there's no need to allow users to mmap the buffer.
460  */
vmcore_alloc_buf(size_t size)461 static inline char *vmcore_alloc_buf(size_t size)
462 {
463 #ifdef CONFIG_MMU
464 	return vmalloc_user(size);
465 #else
466 	return vzalloc(size);
467 #endif
468 }
469 
470 /*
471  * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
472  * essential for mmap_vmcore() in order to map physically
473  * non-contiguous objects (ELF header, ELF note segment and memory
474  * regions in the 1st kernel pointed to by PT_LOAD entries) into
475  * virtually contiguous user-space in ELF layout.
476  */
477 #ifdef CONFIG_MMU
478 /*
479  * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
480  * reported as not being ram with the zero page.
481  *
482  * @vma: vm_area_struct describing requested mapping
483  * @from: start remapping from
484  * @pfn: page frame number to start remapping to
485  * @size: remapping size
486  * @prot: protection bits
487  *
488  * Returns zero on success, -EAGAIN on failure.
489  */
remap_oldmem_pfn_checked(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)490 static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
491 				    unsigned long from, unsigned long pfn,
492 				    unsigned long size, pgprot_t prot)
493 {
494 	unsigned long map_size;
495 	unsigned long pos_start, pos_end, pos;
496 	unsigned long zeropage_pfn = my_zero_pfn(0);
497 	size_t len = 0;
498 
499 	pos_start = pfn;
500 	pos_end = pfn + (size >> PAGE_SHIFT);
501 
502 	for (pos = pos_start; pos < pos_end; ++pos) {
503 		if (!pfn_is_ram(pos)) {
504 			/*
505 			 * We hit a page which is not ram. Remap the continuous
506 			 * region between pos_start and pos-1 and replace
507 			 * the non-ram page at pos with the zero page.
508 			 */
509 			if (pos > pos_start) {
510 				/* Remap continuous region */
511 				map_size = (pos - pos_start) << PAGE_SHIFT;
512 				if (remap_oldmem_pfn_range(vma, from + len,
513 							   pos_start, map_size,
514 							   prot))
515 					goto fail;
516 				len += map_size;
517 			}
518 			/* Remap the zero page */
519 			if (remap_oldmem_pfn_range(vma, from + len,
520 						   zeropage_pfn,
521 						   PAGE_SIZE, prot))
522 				goto fail;
523 			len += PAGE_SIZE;
524 			pos_start = pos + 1;
525 		}
526 	}
527 	if (pos > pos_start) {
528 		/* Remap the rest */
529 		map_size = (pos - pos_start) << PAGE_SHIFT;
530 		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
531 					   map_size, prot))
532 			goto fail;
533 	}
534 	return 0;
535 fail:
536 	do_munmap(vma->vm_mm, from, len, NULL);
537 	return -EAGAIN;
538 }
539 
vmcore_remap_oldmem_pfn(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)540 static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
541 			    unsigned long from, unsigned long pfn,
542 			    unsigned long size, pgprot_t prot)
543 {
544 	/*
545 	 * Check if oldmem_pfn_is_ram was registered to avoid
546 	 * looping over all pages without a reason.
547 	 */
548 	if (oldmem_pfn_is_ram)
549 		return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
550 	else
551 		return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
552 }
553 
mmap_vmcore(struct file * file,struct vm_area_struct * vma)554 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
555 {
556 	size_t size = vma->vm_end - vma->vm_start;
557 	u64 start, end, len, tsz;
558 	struct vmcore *m;
559 
560 	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
561 	end = start + size;
562 
563 	if (size > vmcore_size || end > vmcore_size)
564 		return -EINVAL;
565 
566 	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
567 		return -EPERM;
568 
569 	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
570 	vma->vm_flags |= VM_MIXEDMAP;
571 	vma->vm_ops = &vmcore_mmap_ops;
572 
573 	len = 0;
574 
575 	if (start < elfcorebuf_sz) {
576 		u64 pfn;
577 
578 		tsz = min(elfcorebuf_sz - (size_t)start, size);
579 		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
580 		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
581 				    vma->vm_page_prot))
582 			return -EAGAIN;
583 		size -= tsz;
584 		start += tsz;
585 		len += tsz;
586 
587 		if (size == 0)
588 			return 0;
589 	}
590 
591 	if (start < elfcorebuf_sz + elfnotes_sz) {
592 		void *kaddr;
593 
594 		/* We add device dumps before other elf notes because the
595 		 * other elf notes may not fill the elf notes buffer
596 		 * completely and we will end up with zero-filled data
597 		 * between the elf notes and the device dumps. Tools will
598 		 * then try to decode this zero-filled data as valid notes
599 		 * and we don't want that. Hence, adding device dumps before
600 		 * the other elf notes ensure that zero-filled data can be
601 		 * avoided. This also ensures that the device dumps and
602 		 * other elf notes can be properly mmaped at page aligned
603 		 * address.
604 		 */
605 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
606 		/* Read device dumps */
607 		if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
608 			u64 start_off;
609 
610 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
611 				  (size_t)start, size);
612 			start_off = start - elfcorebuf_sz;
613 			if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
614 						start_off, tsz))
615 				goto fail;
616 
617 			size -= tsz;
618 			start += tsz;
619 			len += tsz;
620 
621 			/* leave now if filled buffer already */
622 			if (!size)
623 				return 0;
624 		}
625 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
626 
627 		/* Read remaining elf notes */
628 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
629 		kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
630 		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
631 						kaddr, 0, tsz))
632 			goto fail;
633 
634 		size -= tsz;
635 		start += tsz;
636 		len += tsz;
637 
638 		if (size == 0)
639 			return 0;
640 	}
641 
642 	list_for_each_entry(m, &vmcore_list, list) {
643 		if (start < m->offset + m->size) {
644 			u64 paddr = 0;
645 
646 			tsz = (size_t)min_t(unsigned long long,
647 					    m->offset + m->size - start, size);
648 			paddr = m->paddr + start - m->offset;
649 			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
650 						    paddr >> PAGE_SHIFT, tsz,
651 						    vma->vm_page_prot))
652 				goto fail;
653 			size -= tsz;
654 			start += tsz;
655 			len += tsz;
656 
657 			if (size == 0)
658 				return 0;
659 		}
660 	}
661 
662 	return 0;
663 fail:
664 	do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
665 	return -EAGAIN;
666 }
667 #else
mmap_vmcore(struct file * file,struct vm_area_struct * vma)668 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
669 {
670 	return -ENOSYS;
671 }
672 #endif
673 
674 static const struct proc_ops vmcore_proc_ops = {
675 	.proc_read	= read_vmcore,
676 	.proc_lseek	= default_llseek,
677 	.proc_mmap	= mmap_vmcore,
678 };
679 
get_new_element(void)680 static struct vmcore* __init get_new_element(void)
681 {
682 	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
683 }
684 
get_vmcore_size(size_t elfsz,size_t elfnotesegsz,struct list_head * vc_list)685 static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
686 			   struct list_head *vc_list)
687 {
688 	u64 size;
689 	struct vmcore *m;
690 
691 	size = elfsz + elfnotesegsz;
692 	list_for_each_entry(m, vc_list, list) {
693 		size += m->size;
694 	}
695 	return size;
696 }
697 
698 /**
699  * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
700  *
701  * @ehdr_ptr: ELF header
702  *
703  * This function updates p_memsz member of each PT_NOTE entry in the
704  * program header table pointed to by @ehdr_ptr to real size of ELF
705  * note segment.
706  */
update_note_header_size_elf64(const Elf64_Ehdr * ehdr_ptr)707 static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
708 {
709 	int i, rc=0;
710 	Elf64_Phdr *phdr_ptr;
711 	Elf64_Nhdr *nhdr_ptr;
712 
713 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
714 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
715 		void *notes_section;
716 		u64 offset, max_sz, sz, real_sz = 0;
717 		if (phdr_ptr->p_type != PT_NOTE)
718 			continue;
719 		max_sz = phdr_ptr->p_memsz;
720 		offset = phdr_ptr->p_offset;
721 		notes_section = kmalloc(max_sz, GFP_KERNEL);
722 		if (!notes_section)
723 			return -ENOMEM;
724 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
725 		if (rc < 0) {
726 			kfree(notes_section);
727 			return rc;
728 		}
729 		nhdr_ptr = notes_section;
730 		while (nhdr_ptr->n_namesz != 0) {
731 			sz = sizeof(Elf64_Nhdr) +
732 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
733 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
734 			if ((real_sz + sz) > max_sz) {
735 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
736 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
737 				break;
738 			}
739 			real_sz += sz;
740 			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
741 		}
742 		kfree(notes_section);
743 		phdr_ptr->p_memsz = real_sz;
744 		if (real_sz == 0) {
745 			pr_warn("Warning: Zero PT_NOTE entries found\n");
746 		}
747 	}
748 
749 	return 0;
750 }
751 
752 /**
753  * get_note_number_and_size_elf64 - get the number of PT_NOTE program
754  * headers and sum of real size of their ELF note segment headers and
755  * data.
756  *
757  * @ehdr_ptr: ELF header
758  * @nr_ptnote: buffer for the number of PT_NOTE program headers
759  * @sz_ptnote: buffer for size of unique PT_NOTE program header
760  *
761  * This function is used to merge multiple PT_NOTE program headers
762  * into a unique single one. The resulting unique entry will have
763  * @sz_ptnote in its phdr->p_mem.
764  *
765  * It is assumed that program headers with PT_NOTE type pointed to by
766  * @ehdr_ptr has already been updated by update_note_header_size_elf64
767  * and each of PT_NOTE program headers has actual ELF note segment
768  * size in its p_memsz member.
769  */
get_note_number_and_size_elf64(const Elf64_Ehdr * ehdr_ptr,int * nr_ptnote,u64 * sz_ptnote)770 static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
771 						 int *nr_ptnote, u64 *sz_ptnote)
772 {
773 	int i;
774 	Elf64_Phdr *phdr_ptr;
775 
776 	*nr_ptnote = *sz_ptnote = 0;
777 
778 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
779 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
780 		if (phdr_ptr->p_type != PT_NOTE)
781 			continue;
782 		*nr_ptnote += 1;
783 		*sz_ptnote += phdr_ptr->p_memsz;
784 	}
785 
786 	return 0;
787 }
788 
789 /**
790  * copy_notes_elf64 - copy ELF note segments in a given buffer
791  *
792  * @ehdr_ptr: ELF header
793  * @notes_buf: buffer into which ELF note segments are copied
794  *
795  * This function is used to copy ELF note segment in the 1st kernel
796  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
797  * size of the buffer @notes_buf is equal to or larger than sum of the
798  * real ELF note segment headers and data.
799  *
800  * It is assumed that program headers with PT_NOTE type pointed to by
801  * @ehdr_ptr has already been updated by update_note_header_size_elf64
802  * and each of PT_NOTE program headers has actual ELF note segment
803  * size in its p_memsz member.
804  */
copy_notes_elf64(const Elf64_Ehdr * ehdr_ptr,char * notes_buf)805 static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
806 {
807 	int i, rc=0;
808 	Elf64_Phdr *phdr_ptr;
809 
810 	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
811 
812 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
813 		u64 offset;
814 		if (phdr_ptr->p_type != PT_NOTE)
815 			continue;
816 		offset = phdr_ptr->p_offset;
817 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
818 					   &offset);
819 		if (rc < 0)
820 			return rc;
821 		notes_buf += phdr_ptr->p_memsz;
822 	}
823 
824 	return 0;
825 }
826 
827 /* Merges all the PT_NOTE headers into one. */
merge_note_headers_elf64(char * elfptr,size_t * elfsz,char ** notes_buf,size_t * notes_sz)828 static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
829 					   char **notes_buf, size_t *notes_sz)
830 {
831 	int i, nr_ptnote=0, rc=0;
832 	char *tmp;
833 	Elf64_Ehdr *ehdr_ptr;
834 	Elf64_Phdr phdr;
835 	u64 phdr_sz = 0, note_off;
836 
837 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
838 
839 	rc = update_note_header_size_elf64(ehdr_ptr);
840 	if (rc < 0)
841 		return rc;
842 
843 	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
844 	if (rc < 0)
845 		return rc;
846 
847 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
848 	*notes_buf = vmcore_alloc_buf(*notes_sz);
849 	if (!*notes_buf)
850 		return -ENOMEM;
851 
852 	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
853 	if (rc < 0)
854 		return rc;
855 
856 	/* Prepare merged PT_NOTE program header. */
857 	phdr.p_type    = PT_NOTE;
858 	phdr.p_flags   = 0;
859 	note_off = sizeof(Elf64_Ehdr) +
860 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
861 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
862 	phdr.p_vaddr   = phdr.p_paddr = 0;
863 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
864 	phdr.p_align   = 0;
865 
866 	/* Add merged PT_NOTE program header*/
867 	tmp = elfptr + sizeof(Elf64_Ehdr);
868 	memcpy(tmp, &phdr, sizeof(phdr));
869 	tmp += sizeof(phdr);
870 
871 	/* Remove unwanted PT_NOTE program headers. */
872 	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
873 	*elfsz = *elfsz - i;
874 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
875 	memset(elfptr + *elfsz, 0, i);
876 	*elfsz = roundup(*elfsz, PAGE_SIZE);
877 
878 	/* Modify e_phnum to reflect merged headers. */
879 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
880 
881 	/* Store the size of all notes.  We need this to update the note
882 	 * header when the device dumps will be added.
883 	 */
884 	elfnotes_orig_sz = phdr.p_memsz;
885 
886 	return 0;
887 }
888 
889 /**
890  * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
891  *
892  * @ehdr_ptr: ELF header
893  *
894  * This function updates p_memsz member of each PT_NOTE entry in the
895  * program header table pointed to by @ehdr_ptr to real size of ELF
896  * note segment.
897  */
update_note_header_size_elf32(const Elf32_Ehdr * ehdr_ptr)898 static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
899 {
900 	int i, rc=0;
901 	Elf32_Phdr *phdr_ptr;
902 	Elf32_Nhdr *nhdr_ptr;
903 
904 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
905 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
906 		void *notes_section;
907 		u64 offset, max_sz, sz, real_sz = 0;
908 		if (phdr_ptr->p_type != PT_NOTE)
909 			continue;
910 		max_sz = phdr_ptr->p_memsz;
911 		offset = phdr_ptr->p_offset;
912 		notes_section = kmalloc(max_sz, GFP_KERNEL);
913 		if (!notes_section)
914 			return -ENOMEM;
915 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
916 		if (rc < 0) {
917 			kfree(notes_section);
918 			return rc;
919 		}
920 		nhdr_ptr = notes_section;
921 		while (nhdr_ptr->n_namesz != 0) {
922 			sz = sizeof(Elf32_Nhdr) +
923 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
924 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
925 			if ((real_sz + sz) > max_sz) {
926 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
927 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
928 				break;
929 			}
930 			real_sz += sz;
931 			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
932 		}
933 		kfree(notes_section);
934 		phdr_ptr->p_memsz = real_sz;
935 		if (real_sz == 0) {
936 			pr_warn("Warning: Zero PT_NOTE entries found\n");
937 		}
938 	}
939 
940 	return 0;
941 }
942 
943 /**
944  * get_note_number_and_size_elf32 - get the number of PT_NOTE program
945  * headers and sum of real size of their ELF note segment headers and
946  * data.
947  *
948  * @ehdr_ptr: ELF header
949  * @nr_ptnote: buffer for the number of PT_NOTE program headers
950  * @sz_ptnote: buffer for size of unique PT_NOTE program header
951  *
952  * This function is used to merge multiple PT_NOTE program headers
953  * into a unique single one. The resulting unique entry will have
954  * @sz_ptnote in its phdr->p_mem.
955  *
956  * It is assumed that program headers with PT_NOTE type pointed to by
957  * @ehdr_ptr has already been updated by update_note_header_size_elf32
958  * and each of PT_NOTE program headers has actual ELF note segment
959  * size in its p_memsz member.
960  */
get_note_number_and_size_elf32(const Elf32_Ehdr * ehdr_ptr,int * nr_ptnote,u64 * sz_ptnote)961 static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
962 						 int *nr_ptnote, u64 *sz_ptnote)
963 {
964 	int i;
965 	Elf32_Phdr *phdr_ptr;
966 
967 	*nr_ptnote = *sz_ptnote = 0;
968 
969 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
970 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
971 		if (phdr_ptr->p_type != PT_NOTE)
972 			continue;
973 		*nr_ptnote += 1;
974 		*sz_ptnote += phdr_ptr->p_memsz;
975 	}
976 
977 	return 0;
978 }
979 
980 /**
981  * copy_notes_elf32 - copy ELF note segments in a given buffer
982  *
983  * @ehdr_ptr: ELF header
984  * @notes_buf: buffer into which ELF note segments are copied
985  *
986  * This function is used to copy ELF note segment in the 1st kernel
987  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
988  * size of the buffer @notes_buf is equal to or larger than sum of the
989  * real ELF note segment headers and data.
990  *
991  * It is assumed that program headers with PT_NOTE type pointed to by
992  * @ehdr_ptr has already been updated by update_note_header_size_elf32
993  * and each of PT_NOTE program headers has actual ELF note segment
994  * size in its p_memsz member.
995  */
copy_notes_elf32(const Elf32_Ehdr * ehdr_ptr,char * notes_buf)996 static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
997 {
998 	int i, rc=0;
999 	Elf32_Phdr *phdr_ptr;
1000 
1001 	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
1002 
1003 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1004 		u64 offset;
1005 		if (phdr_ptr->p_type != PT_NOTE)
1006 			continue;
1007 		offset = phdr_ptr->p_offset;
1008 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1009 					   &offset);
1010 		if (rc < 0)
1011 			return rc;
1012 		notes_buf += phdr_ptr->p_memsz;
1013 	}
1014 
1015 	return 0;
1016 }
1017 
1018 /* Merges all the PT_NOTE headers into one. */
merge_note_headers_elf32(char * elfptr,size_t * elfsz,char ** notes_buf,size_t * notes_sz)1019 static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1020 					   char **notes_buf, size_t *notes_sz)
1021 {
1022 	int i, nr_ptnote=0, rc=0;
1023 	char *tmp;
1024 	Elf32_Ehdr *ehdr_ptr;
1025 	Elf32_Phdr phdr;
1026 	u64 phdr_sz = 0, note_off;
1027 
1028 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1029 
1030 	rc = update_note_header_size_elf32(ehdr_ptr);
1031 	if (rc < 0)
1032 		return rc;
1033 
1034 	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1035 	if (rc < 0)
1036 		return rc;
1037 
1038 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
1039 	*notes_buf = vmcore_alloc_buf(*notes_sz);
1040 	if (!*notes_buf)
1041 		return -ENOMEM;
1042 
1043 	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1044 	if (rc < 0)
1045 		return rc;
1046 
1047 	/* Prepare merged PT_NOTE program header. */
1048 	phdr.p_type    = PT_NOTE;
1049 	phdr.p_flags   = 0;
1050 	note_off = sizeof(Elf32_Ehdr) +
1051 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1052 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
1053 	phdr.p_vaddr   = phdr.p_paddr = 0;
1054 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
1055 	phdr.p_align   = 0;
1056 
1057 	/* Add merged PT_NOTE program header*/
1058 	tmp = elfptr + sizeof(Elf32_Ehdr);
1059 	memcpy(tmp, &phdr, sizeof(phdr));
1060 	tmp += sizeof(phdr);
1061 
1062 	/* Remove unwanted PT_NOTE program headers. */
1063 	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1064 	*elfsz = *elfsz - i;
1065 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1066 	memset(elfptr + *elfsz, 0, i);
1067 	*elfsz = roundup(*elfsz, PAGE_SIZE);
1068 
1069 	/* Modify e_phnum to reflect merged headers. */
1070 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1071 
1072 	/* Store the size of all notes.  We need this to update the note
1073 	 * header when the device dumps will be added.
1074 	 */
1075 	elfnotes_orig_sz = phdr.p_memsz;
1076 
1077 	return 0;
1078 }
1079 
1080 /* Add memory chunks represented by program headers to vmcore list. Also update
1081  * the new offset fields of exported program headers. */
process_ptload_program_headers_elf64(char * elfptr,size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1082 static int __init process_ptload_program_headers_elf64(char *elfptr,
1083 						size_t elfsz,
1084 						size_t elfnotes_sz,
1085 						struct list_head *vc_list)
1086 {
1087 	int i;
1088 	Elf64_Ehdr *ehdr_ptr;
1089 	Elf64_Phdr *phdr_ptr;
1090 	loff_t vmcore_off;
1091 	struct vmcore *new;
1092 
1093 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
1094 	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1095 
1096 	/* Skip Elf header, program headers and Elf note segment. */
1097 	vmcore_off = elfsz + elfnotes_sz;
1098 
1099 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1100 		u64 paddr, start, end, size;
1101 
1102 		if (phdr_ptr->p_type != PT_LOAD)
1103 			continue;
1104 
1105 		paddr = phdr_ptr->p_offset;
1106 		start = rounddown(paddr, PAGE_SIZE);
1107 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1108 		size = end - start;
1109 
1110 		/* Add this contiguous chunk of memory to vmcore list.*/
1111 		new = get_new_element();
1112 		if (!new)
1113 			return -ENOMEM;
1114 		new->paddr = start;
1115 		new->size = size;
1116 		list_add_tail(&new->list, vc_list);
1117 
1118 		/* Update the program header offset. */
1119 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1120 		vmcore_off = vmcore_off + size;
1121 	}
1122 	return 0;
1123 }
1124 
process_ptload_program_headers_elf32(char * elfptr,size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1125 static int __init process_ptload_program_headers_elf32(char *elfptr,
1126 						size_t elfsz,
1127 						size_t elfnotes_sz,
1128 						struct list_head *vc_list)
1129 {
1130 	int i;
1131 	Elf32_Ehdr *ehdr_ptr;
1132 	Elf32_Phdr *phdr_ptr;
1133 	loff_t vmcore_off;
1134 	struct vmcore *new;
1135 
1136 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1137 	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1138 
1139 	/* Skip Elf header, program headers and Elf note segment. */
1140 	vmcore_off = elfsz + elfnotes_sz;
1141 
1142 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1143 		u64 paddr, start, end, size;
1144 
1145 		if (phdr_ptr->p_type != PT_LOAD)
1146 			continue;
1147 
1148 		paddr = phdr_ptr->p_offset;
1149 		start = rounddown(paddr, PAGE_SIZE);
1150 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1151 		size = end - start;
1152 
1153 		/* Add this contiguous chunk of memory to vmcore list.*/
1154 		new = get_new_element();
1155 		if (!new)
1156 			return -ENOMEM;
1157 		new->paddr = start;
1158 		new->size = size;
1159 		list_add_tail(&new->list, vc_list);
1160 
1161 		/* Update the program header offset */
1162 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1163 		vmcore_off = vmcore_off + size;
1164 	}
1165 	return 0;
1166 }
1167 
1168 /* Sets offset fields of vmcore elements. */
set_vmcore_list_offsets(size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1169 static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1170 				    struct list_head *vc_list)
1171 {
1172 	loff_t vmcore_off;
1173 	struct vmcore *m;
1174 
1175 	/* Skip Elf header, program headers and Elf note segment. */
1176 	vmcore_off = elfsz + elfnotes_sz;
1177 
1178 	list_for_each_entry(m, vc_list, list) {
1179 		m->offset = vmcore_off;
1180 		vmcore_off += m->size;
1181 	}
1182 }
1183 
free_elfcorebuf(void)1184 static void free_elfcorebuf(void)
1185 {
1186 	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1187 	elfcorebuf = NULL;
1188 	vfree(elfnotes_buf);
1189 	elfnotes_buf = NULL;
1190 }
1191 
parse_crash_elf64_headers(void)1192 static int __init parse_crash_elf64_headers(void)
1193 {
1194 	int rc=0;
1195 	Elf64_Ehdr ehdr;
1196 	u64 addr;
1197 
1198 	addr = elfcorehdr_addr;
1199 
1200 	/* Read Elf header */
1201 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1202 	if (rc < 0)
1203 		return rc;
1204 
1205 	/* Do some basic Verification. */
1206 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1207 		(ehdr.e_type != ET_CORE) ||
1208 		!vmcore_elf64_check_arch(&ehdr) ||
1209 		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1210 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1211 		ehdr.e_version != EV_CURRENT ||
1212 		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1213 		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1214 		ehdr.e_phnum == 0) {
1215 		pr_warn("Warning: Core image elf header is not sane\n");
1216 		return -EINVAL;
1217 	}
1218 
1219 	/* Read in all elf headers. */
1220 	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1221 				ehdr.e_phnum * sizeof(Elf64_Phdr);
1222 	elfcorebuf_sz = elfcorebuf_sz_orig;
1223 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1224 					      get_order(elfcorebuf_sz_orig));
1225 	if (!elfcorebuf)
1226 		return -ENOMEM;
1227 	addr = elfcorehdr_addr;
1228 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1229 	if (rc < 0)
1230 		goto fail;
1231 
1232 	/* Merge all PT_NOTE headers into one. */
1233 	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1234 				      &elfnotes_buf, &elfnotes_sz);
1235 	if (rc)
1236 		goto fail;
1237 	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1238 						  elfnotes_sz, &vmcore_list);
1239 	if (rc)
1240 		goto fail;
1241 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1242 	return 0;
1243 fail:
1244 	free_elfcorebuf();
1245 	return rc;
1246 }
1247 
parse_crash_elf32_headers(void)1248 static int __init parse_crash_elf32_headers(void)
1249 {
1250 	int rc=0;
1251 	Elf32_Ehdr ehdr;
1252 	u64 addr;
1253 
1254 	addr = elfcorehdr_addr;
1255 
1256 	/* Read Elf header */
1257 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1258 	if (rc < 0)
1259 		return rc;
1260 
1261 	/* Do some basic Verification. */
1262 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1263 		(ehdr.e_type != ET_CORE) ||
1264 		!vmcore_elf32_check_arch(&ehdr) ||
1265 		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1266 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1267 		ehdr.e_version != EV_CURRENT ||
1268 		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1269 		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1270 		ehdr.e_phnum == 0) {
1271 		pr_warn("Warning: Core image elf header is not sane\n");
1272 		return -EINVAL;
1273 	}
1274 
1275 	/* Read in all elf headers. */
1276 	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1277 	elfcorebuf_sz = elfcorebuf_sz_orig;
1278 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1279 					      get_order(elfcorebuf_sz_orig));
1280 	if (!elfcorebuf)
1281 		return -ENOMEM;
1282 	addr = elfcorehdr_addr;
1283 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1284 	if (rc < 0)
1285 		goto fail;
1286 
1287 	/* Merge all PT_NOTE headers into one. */
1288 	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1289 				      &elfnotes_buf, &elfnotes_sz);
1290 	if (rc)
1291 		goto fail;
1292 	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1293 						  elfnotes_sz, &vmcore_list);
1294 	if (rc)
1295 		goto fail;
1296 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1297 	return 0;
1298 fail:
1299 	free_elfcorebuf();
1300 	return rc;
1301 }
1302 
parse_crash_elf_headers(void)1303 static int __init parse_crash_elf_headers(void)
1304 {
1305 	unsigned char e_ident[EI_NIDENT];
1306 	u64 addr;
1307 	int rc=0;
1308 
1309 	addr = elfcorehdr_addr;
1310 	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1311 	if (rc < 0)
1312 		return rc;
1313 	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1314 		pr_warn("Warning: Core image elf header not found\n");
1315 		return -EINVAL;
1316 	}
1317 
1318 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1319 		rc = parse_crash_elf64_headers();
1320 		if (rc)
1321 			return rc;
1322 	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1323 		rc = parse_crash_elf32_headers();
1324 		if (rc)
1325 			return rc;
1326 	} else {
1327 		pr_warn("Warning: Core image elf header is not sane\n");
1328 		return -EINVAL;
1329 	}
1330 
1331 	/* Determine vmcore size. */
1332 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1333 				      &vmcore_list);
1334 
1335 	return 0;
1336 }
1337 
1338 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1339 /**
1340  * vmcoredd_write_header - Write vmcore device dump header at the
1341  * beginning of the dump's buffer.
1342  * @buf: Output buffer where the note is written
1343  * @data: Dump info
1344  * @size: Size of the dump
1345  *
1346  * Fills beginning of the dump's buffer with vmcore device dump header.
1347  */
vmcoredd_write_header(void * buf,struct vmcoredd_data * data,u32 size)1348 static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1349 				  u32 size)
1350 {
1351 	struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1352 
1353 	vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1354 	vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1355 	vdd_hdr->n_type = NT_VMCOREDD;
1356 
1357 	strncpy((char *)vdd_hdr->name, VMCOREDD_NOTE_NAME,
1358 		sizeof(vdd_hdr->name));
1359 	memcpy(vdd_hdr->dump_name, data->dump_name, sizeof(vdd_hdr->dump_name));
1360 }
1361 
1362 /**
1363  * vmcoredd_update_program_headers - Update all Elf program headers
1364  * @elfptr: Pointer to elf header
1365  * @elfnotesz: Size of elf notes aligned to page size
1366  * @vmcoreddsz: Size of device dumps to be added to elf note header
1367  *
1368  * Determine type of Elf header (Elf64 or Elf32) and update the elf note size.
1369  * Also update the offsets of all the program headers after the elf note header.
1370  */
vmcoredd_update_program_headers(char * elfptr,size_t elfnotesz,size_t vmcoreddsz)1371 static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1372 					    size_t vmcoreddsz)
1373 {
1374 	unsigned char *e_ident = (unsigned char *)elfptr;
1375 	u64 start, end, size;
1376 	loff_t vmcore_off;
1377 	u32 i;
1378 
1379 	vmcore_off = elfcorebuf_sz + elfnotesz;
1380 
1381 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1382 		Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1383 		Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1384 
1385 		/* Update all program headers */
1386 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1387 			if (phdr->p_type == PT_NOTE) {
1388 				/* Update note size */
1389 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1390 				phdr->p_filesz = phdr->p_memsz;
1391 				continue;
1392 			}
1393 
1394 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1395 			end = roundup(phdr->p_offset + phdr->p_memsz,
1396 				      PAGE_SIZE);
1397 			size = end - start;
1398 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1399 			vmcore_off += size;
1400 		}
1401 	} else {
1402 		Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1403 		Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1404 
1405 		/* Update all program headers */
1406 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1407 			if (phdr->p_type == PT_NOTE) {
1408 				/* Update note size */
1409 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1410 				phdr->p_filesz = phdr->p_memsz;
1411 				continue;
1412 			}
1413 
1414 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1415 			end = roundup(phdr->p_offset + phdr->p_memsz,
1416 				      PAGE_SIZE);
1417 			size = end - start;
1418 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1419 			vmcore_off += size;
1420 		}
1421 	}
1422 }
1423 
1424 /**
1425  * vmcoredd_update_size - Update the total size of the device dumps and update
1426  * Elf header
1427  * @dump_size: Size of the current device dump to be added to total size
1428  *
1429  * Update the total size of all the device dumps and update the Elf program
1430  * headers. Calculate the new offsets for the vmcore list and update the
1431  * total vmcore size.
1432  */
vmcoredd_update_size(size_t dump_size)1433 static void vmcoredd_update_size(size_t dump_size)
1434 {
1435 	vmcoredd_orig_sz += dump_size;
1436 	elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1437 	vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1438 					vmcoredd_orig_sz);
1439 
1440 	/* Update vmcore list offsets */
1441 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1442 
1443 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1444 				      &vmcore_list);
1445 	proc_vmcore->size = vmcore_size;
1446 }
1447 
1448 /**
1449  * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1450  * @data: dump info.
1451  *
1452  * Allocate a buffer and invoke the calling driver's dump collect routine.
1453  * Write Elf note at the beginning of the buffer to indicate vmcore device
1454  * dump and add the dump to global list.
1455  */
vmcore_add_device_dump(struct vmcoredd_data * data)1456 int vmcore_add_device_dump(struct vmcoredd_data *data)
1457 {
1458 	struct vmcoredd_node *dump;
1459 	void *buf = NULL;
1460 	size_t data_size;
1461 	int ret;
1462 
1463 	if (vmcoredd_disabled) {
1464 		pr_err_once("Device dump is disabled\n");
1465 		return -EINVAL;
1466 	}
1467 
1468 	if (!data || !strlen(data->dump_name) ||
1469 	    !data->vmcoredd_callback || !data->size)
1470 		return -EINVAL;
1471 
1472 	dump = vzalloc(sizeof(*dump));
1473 	if (!dump) {
1474 		ret = -ENOMEM;
1475 		goto out_err;
1476 	}
1477 
1478 	/* Keep size of the buffer page aligned so that it can be mmaped */
1479 	data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1480 			    PAGE_SIZE);
1481 
1482 	/* Allocate buffer for driver's to write their dumps */
1483 	buf = vmcore_alloc_buf(data_size);
1484 	if (!buf) {
1485 		ret = -ENOMEM;
1486 		goto out_err;
1487 	}
1488 
1489 	vmcoredd_write_header(buf, data, data_size -
1490 			      sizeof(struct vmcoredd_header));
1491 
1492 	/* Invoke the driver's dump collection routing */
1493 	ret = data->vmcoredd_callback(data, buf +
1494 				      sizeof(struct vmcoredd_header));
1495 	if (ret)
1496 		goto out_err;
1497 
1498 	dump->buf = buf;
1499 	dump->size = data_size;
1500 
1501 	/* Add the dump to driver sysfs list */
1502 	mutex_lock(&vmcoredd_mutex);
1503 	list_add_tail(&dump->list, &vmcoredd_list);
1504 	mutex_unlock(&vmcoredd_mutex);
1505 
1506 	vmcoredd_update_size(data_size);
1507 	return 0;
1508 
1509 out_err:
1510 	if (buf)
1511 		vfree(buf);
1512 
1513 	if (dump)
1514 		vfree(dump);
1515 
1516 	return ret;
1517 }
1518 EXPORT_SYMBOL(vmcore_add_device_dump);
1519 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1520 
1521 /* Free all dumps in vmcore device dump list */
vmcore_free_device_dumps(void)1522 static void vmcore_free_device_dumps(void)
1523 {
1524 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1525 	mutex_lock(&vmcoredd_mutex);
1526 	while (!list_empty(&vmcoredd_list)) {
1527 		struct vmcoredd_node *dump;
1528 
1529 		dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1530 					list);
1531 		list_del(&dump->list);
1532 		vfree(dump->buf);
1533 		vfree(dump);
1534 	}
1535 	mutex_unlock(&vmcoredd_mutex);
1536 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1537 }
1538 
1539 /* Init function for vmcore module. */
vmcore_init(void)1540 static int __init vmcore_init(void)
1541 {
1542 	int rc = 0;
1543 
1544 	/* Allow architectures to allocate ELF header in 2nd kernel */
1545 	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1546 	if (rc)
1547 		return rc;
1548 	/*
1549 	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1550 	 * then capture the dump.
1551 	 */
1552 	if (!(is_vmcore_usable()))
1553 		return rc;
1554 	rc = parse_crash_elf_headers();
1555 	if (rc) {
1556 		pr_warn("Kdump: vmcore not initialized\n");
1557 		return rc;
1558 	}
1559 	elfcorehdr_free(elfcorehdr_addr);
1560 	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1561 
1562 	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1563 	if (proc_vmcore)
1564 		proc_vmcore->size = vmcore_size;
1565 	return 0;
1566 }
1567 fs_initcall(vmcore_init);
1568 
1569 /* Cleanup function for vmcore module. */
vmcore_cleanup(void)1570 void vmcore_cleanup(void)
1571 {
1572 	if (proc_vmcore) {
1573 		proc_remove(proc_vmcore);
1574 		proc_vmcore = NULL;
1575 	}
1576 
1577 	/* clear the vmcore list. */
1578 	while (!list_empty(&vmcore_list)) {
1579 		struct vmcore *m;
1580 
1581 		m = list_first_entry(&vmcore_list, struct vmcore, list);
1582 		list_del(&m->list);
1583 		kfree(m);
1584 	}
1585 	free_elfcorebuf();
1586 
1587 	/* clear vmcore device dump list */
1588 	vmcore_free_device_dumps();
1589 }
1590